Modelling irregularly sampled profiles of nonnegative dog triglyceride responses under different distributional assumptions

Abstract

General methodology for modelling series of non-negative data observed at unequally spaced times is developed. The parameterization enables both the importance of the serial association, as well the order of this dependence to be expressed. An example is given where the effects of three fibre based diets on dog triglyceride profiles are analysed and compared. Many different types of models based on common distributions such as the normal, exponential, gamma, Weibull and log-normal observations are presented. Comparison of possibly non-nested models fitted on the same data set is made using the Akaike criterion.